HSC Section 3 - Trauma, Critical Care and Sleep Medicine

Woodson et al

the month 60 visit had higher AHI, higher ODI, and lower FOSQ scores at baseline. These differences disappeared when evaluation was performed at 12 months while therapy was activated. 1 Among the 71 participants who voluntarily completed a 60-month PSG study, age, AHI, and BMI base- line parameters were not significantly different from those who did not complete the PSG, the 97 patients meeting pro- tocol, or the original cohort ( Table 1 ). The AHI treatment response did differ at 12 months (74% vs 52%, P \ .002), but AHI treatment response at 18 and 36 months, change in sleepiness, and change in quality of life did not differ between groups that did and did not complete the 60-month PSG. Primary Outcome Measures The efficacy measures of AHI and ODI decreased from baseline to the 12-month assessment and remained stable at 36 and 60 months ( Table 2 , Figure 2 ). A decrease in AHI . 50% and an AHI \ 20, which were the a priori definition of success, were observed in 75% of participants with 5- year PSG ( Figure 2 ). 12 Forty-four percent and 78% of par- ticipants had AHIs \ 5 and \ 15 at 5-year PSG, respectively. Given the number lost to follow-up over the extended follow-up, an LOCF analysis from the cohort at 12, 18, or 36 months was performed. LOCF demonstrated a mean AHI at 5 years of 15.1 6 1.5, with a median of 7.6 and a response rate of 63% (5 deaths and 3 explants were counted as nonresponders), which was similar to the responder rate of 66% at 12 months. Based on the LOCF and multiple- imputation methods to account for missing data, the change of AHI from baseline was similar at 36 and 60 months and did not change with different sensitivity analysis ( Table 3 ). In addition to sensitivity analysis with the LOCF and multiple-imputation methods, we conducted best- and worst-case analyses, in which the minimal and maximal values from available postoperative AHI at 12, 18, and 36 months were used for all patients who did not complete the 60-month PSG. In the best-case analysis, the mean AHI was 12.3 6 15.4, with a change of 2 19.8 6 15.8 (95% CI, 2 22.5 to 2 17.0) at 60 months from baseline. The worst- case analysis demonstrated a mean AHI of 17.0 6 18.2, with a change of 2 15.0 6 16.6 (95% CI, 2 17.9 to 2 12.1) at 60 months from baseline. Changes from baseline in best- and worst-case analyses were not significantly different. When the 5-year AHI responders and nonresponders were compared, univariate analysis demonstrated differ- ences in age and baseline ODI between groups. A multivari- able stepwise regression analysis including age, BMI, sex, neck circumference, prior uvulopalatopharyngoplasty, and baseline AHI, ODI, FOSQ, and ESS demonstrated that only a lower ODI was predictive of 5-year AHI responders ( Table 4 ). Self-reported Outcome Measures FOSQ and ESS improvements observed at prior evaluation periods persisted at 5 years. The average increase of FOSQ was 3.2 units, as observed and unchanged with the

Figure 1. Study flow is shown over 5 years until study completion. Included patients underwent protocol evaluation and follow-up. Nonincluded patients with description are on the right.

analyses included last observation carried forward (LOCF), repeated measures regression, multiple imputation, and maximum likelihood estimation. 11 The LOCF analysis imputed the last available follow-up value for any missing data at months 36 and 60. The repeated measures analysis included all available baseline and follow-up data in a repeated measures regression model and provided least squares estimates of the means at 36 and 60 months. The multiple-imputation analysis created 10 imputed data sets for each parameter, with all available baseline and follow- up data used as predictors. The means at months 36 and 60 were estimated within each imputed data set and combined across imputations. The maximum likelihood estimation analysis provided estimates for the outcomes at months 36 and 60, which maximizes the probability of the observed data. A stepwise multivariable logistic model was used to determine key baseline factors associated with therapy response. Analyses were performed with SAS 9.2 software (SAS Institute). Of the 126 participants who underwent implantation, 97 (78%) completed the 5-year follow-up visit ( Figure 1 ). Among the 29 participants who did not complete the 5-year assessment, 21 were lost to follow-up within the prespeci- fied time frame; 5 died of unrelated causes (sudden death, cardiac arrest after a fall and blunt chest trauma, homicide, malignant melanoma, and myelodysplastic syndrome); and 3 had the device explanted (implantable pulse generator removal in a nonresponder, system removal in a nonrespon- der, and nonelective removal in a responder due to septic arthritis). Of the 97 patients meeting the 5-year follow-up protocol, 71 volunteered for an overnight in-laboratory poly- somnographic evaluation. The mean 6 SD BMI at 5 years was 28.6 6 2.8, unchanged from baseline. Among the 97 participants who did complete the protocol as compared with those who did not, baseline age, BMI, and ESS were similar, but subjects who did not complete Results Participants

184